Coastline Extraction in Vhr Imagery Using Mathematical Morphology with Spatial and Spectral Knowledge

In this article, we are dealing with the problem of coastline extraction in Very High Resolution (VHR) multispectral images (Quickbird) on the Normandy Coast (France). Locating precisely the coastline is a crucial task in the context of coastal resource management and planning. In VHR imagery, some details on coastal zone become visible and the coastline definition depends on the geomorphologic context. According to the type of coastal units (sandy beach, wetlands, dune, cliff), several definitions for the coastline has to be used. So in this paper we propose a new approach in two steps based on morphological tools to extract coastline according to their context. More precisely, we first perform two detections of possible coastline pixels (respectively without false positive and without false negative). To do so, we apply a recent extension to multivariate images of the hit-or-miss transform, the morphological template matching tool, and rely on expert knowledge to define the sought templates. We then combine these two results through a double thresholding procedure followed by a final marker-based watershed to locate the exact coastline. In order to assess the performance and reliability of our method, results are compared with some ground-truth given by expert visual analysis. This comparison is made both visually and quantitatively. Results show the high performance of our method and its relevance to the problem under consideration.